Why this appendix exists
This field generates vocabulary faster than it generates evidence. AEO, GEO, LLMO and AIO are sold as four disciplines and are largely one. "Agentic" is applied to a scripted chatbot and to a system that transacts under a mandate.
The cost is not aesthetic. When you and a vendor use the same word for different things, you buy something other than what you agreed. This appendix fixes the meanings this course uses, marks where the market disagrees, and is the vocabulary the final exam draws on.
A. Discovery and visibility
| Term | Definition |
|---|---|
| Generative engine | A system that answers a question with synthesised text and a small number of cited sources, rather than a ranked list |
| AI Overview / AI Mode | Google's generative answer formats within Search; distinct surfaces with distinct behaviour |
| GEO / AEO / LLMO / AIO | Four names for substantially one practice: making content and data more likely to be retrieved, used and cited by generative engines. This course uses LLM visibility for the outcome and GEO for the practice, following the peer-reviewed usage |
| Citation | A source attributed under or within a generative answer. The unit of visibility that matters |
| Mention | Your brand named in the answer text, with or without a citation. Weaker than a citation, and not nothing |
| Share of voice (generative) | Your mention or citation rate across a fixed prompt set, relative to named competitors |
| Prompt set | A fixed, versioned list of buying questions run on a schedule against chosen engines. The measurement instrument of Modules 3, 5 and 6 |
| Mention rate | Prompts in your set where your brand appears in the answer text, over total prompts. Module 3 |
| Owned-citation share | Citations pointing at properties you control, over all citations in those answers. Not the same as mention rate, and the distinction is load-bearing |
| Citation source register | The ranked list of sources grounding answers about you, tagged by ring — owned, claimed, earned, ambient — with an owner per row. Module 5 |
| Entity layer | The identity facts and sameAs assertions that let an engine know the things bearing your name are one company |
| Query fan-out | The engine's decomposition of one question into several retrieval queries. You compete against queries you never see |
| Retrieval | Fetching candidate passages for the model to work from. Roughly what "ranking" used to buy you |
| Grounding | Placing retrieved material in the model's context so the answer is built from it. Being retrieved does not guarantee being grounded |
| Hallucination | Confident output that is not true. In commerce it most often appears as a wrong price, stock status or returns policy |
| llms.txt | A proposed plain-text convention describing a site for language models. Not a ratified standard; support is inconsistent. Cheap to publish, not a strategy |
B. Machine legibility
| Term | Definition |
|---|---|
| Machine-readability | Whether a machine can extract the commercially relevant facts from your pages and data without executing your interface |
| Product feed | Structured product data syndicated to platforms. On several surfaces it is not an export but the interface |
| Feed freshness | The maximum age of price and availability visible on external surfaces. Now a discovery input, not only an operational metric |
| Structured data / schema markup | Machine-parsable statements in a page about what it describes — Product, Offer, MerchantReturnPolicy and others |
| GTIN / MPN | Global trade item number and manufacturer part number. How a machine knows your product is the same product it saw elsewhere |
| Controlled vocabulary | A fixed value list for an attribute — a colour family alongside forty marketing colour names |
| Negative attribute | An explicit statement of what a product is not suitable for. Prevents mismatched recommendations, and therefore returns |
| Server-side rendering | Facts present in the HTML a machine receives, rather than assembled later by script. The difference between visible and invisible to many crawlers |
| Constraint | What a shopper actually specifies — washable, fits 60cm, safe for pets. The unit a machine matches on, and usually absent from taxonomies built for navigation |
C. Agents and protocols
| Term | Definition |
|---|---|
| Shopping agent | Software that discovers, compares and sometimes transacts on a person's behalf under a mandate |
| Mandate | The instruction and limits a person gives an agent — budget, preferences, constraints. Usually invisible to the merchant, and the decisive evidence in a dispute |
| AI-influenced purchase | A human buys after consulting a generative engine. The large majority of AI-affected revenue today |
| Agent-executed purchase | Software completes the transaction. Small volume, long lead time to support |
| ACP — Agentic Commerce Protocol | Open standard for agent-initiated checkout published by OpenAI and Stripe, September 2025, Apache 2.0 |
| UCP — Universal Commerce Protocol | Google's open standard announced January 2026, covering discovery through post-purchase across its surfaces, co-developed with major retailers |
| Agent-scoped credential | A payment credential issued for an agent and limited in scope, so agent transactions are distinguishable at the network level |
| Idempotency | The guarantee that a retried request does not create a second order. Unremarkable until a machine retries |
| Web Bot Auth | Cryptographic identification of a crawler or agent, so a legitimate one can be told from something using its name |
| Pay per crawl | Charging automated clients for access to content rather than only allowing or blocking. Cloudflare's term for the model, and now the general one |
| AP2 — Agent Payments Protocol | Google's payment-layer specification, September 2025, contributing the cryptographically signed mandate record |
| SCA — strong customer authentication | The PSD2 requirement that applies to European transactions regardless of who initiated them. No protocol removes it |
| EU AI Act Article 50 | The transparency obligation, applicable since 2 August 2026, that a system interacting with people must disclose that it is AI |
| Prompt injection | Text placed in content an agent reads, crafted to be treated as instruction rather than as data. First on the OWASP LLM risk list, and with no reliable general defence |
| Access policy | Your deliberate decision about which machines may fetch what. In most organisations, currently an inherited CDN default |
D. Measurement and commerce
| Term | Definition |
|---|---|
| Referrer loss | Visits arriving without a usable source, misfiled as direct. A principal cause of AI undercounting |
| Floor and ceiling | Reporting measured and triangulated bounds instead of a single false-precision figure |
| Leading indicator | A measurement that moves before revenue does. Here, mention rate and owned-citation share |
| Counterfactual | What would have happened anyway. Naming it is what makes an improvement claim survivable |
| Pre-registration | Fixing metric, baseline and decision rule before the work starts |
| Revenue per visit | Revenue divided by sessions. More robust than conversion rate when traffic mix is shifting |
| False decline | A legitimate transaction refused by risk controls. The likely first cost of agent traffic |
| Evidence pack | The record assembled to defend a dispute: identity, mandate, what was presented, authorisation, timestamps, fulfilment |
| Retail media | Advertising sold by a retailer against its own audience — increasingly including its AI surfaces |
| Conversational placement | Paid inventory inside an assistant experience. Verify it is billable with reporting before it enters a plan |
E. Words this course uses carefully
"Agentic." Applied to everything from a rules-based chatbot to a system transacting under a mandate. Ask which of the two is meant, every time.
"AI traffic." Reported as a channel; in reality a floor with four known leaks. Prefer AI-influenced demand with stated bounds.
"Optimised for AI." Almost always means schema was added. Ask which of the four legibility layers was worked on and what the pass criterion was.
"Partner." In this market, frequently means "appeared on an announcement slide." Ask whether they are live, in your market, in your category.
Self-check
- A vendor offers "AEO services." What do you ask them to distinguish it from GEO, LLMO and ordinary structured-data work?
- Explain the difference between retrieval and grounding, and why being retrieved is not enough.
- What is a mandate, who holds the record of it, and why does that matter in a dispute?
- Why is AI-influenced demand a more defensible reporting term than AI traffic?
- Give an example of a negative attribute in your own catalogue that would prevent a mismatched recommendation.